Update docker-compose.yml

[BOT9315]

It publishes ports 3000:3000, 3001:3001, and 5005:5005/udp GitHub
The entrypoint starts the server with --bind-addr 0.0.0.0, --http-port 3000, --ws-port 3001
When RUVIEW_API_TOKEN is empty, /api/v1/* auth is a no-op, and /ws/sensing is unauthenticated even when a token IS set
The suggested fix is to fail startup when binding to 0.0.0.0 or publishing ports without a non-empty token, and to bind to loopback by default, requiring an explicit unsafe/LAN flag to expose it wider
A separate audit also flagged missing resource limits (no mem_limit/cpus/pids_limit) and no read_only filesystem restriction on the same file

[PrinceAudre]

The hardening intent is good (loopback-bind by default, required RUVIEW_API_TOKEN, no-new-privileges, resource limits) — those parts are worth keeping. But a few changes here will break docker compose up before they help:

1. Broken healthcheck. curl -f http://localhost:3000/api/v1/health targets a route that doesn't exist — the server registers /health, /health/live, /health/ready (main.rs ~7992), nothing under /api/v1/health. So curl -f gets a 404 → non-zero → the container is permanently unhealthy, and anything with depends_on: service_healthy stalls. Change the path to /health.
2. Drops the local build + renames the image. Replaces build: docker/Dockerfile.rust with image: ${RUVIEW_IMAGE:-ghcr.io/ruvnet/ruview:latest} — a different name from the published ruvnet/wifi-densepose:latest, and docker compose up --build no longer works. Please restore build: (or confirm the ghcr image is published under that name).
3. Deletes the python-sensing service (ports 8765/8080) — that's a functional removal, not hardening.
4. read_only: true + dropping MODELS_DIR leaves only /tmp writable, so model loading and /api/v1/recording/* writes to data/models fail for real (non-simulated) use. Add a writable volume/tmpfs for those dirs, or drop read_only.

Separately: #864's underlying concern (the /ws/sensing stream sits outside the /api/v1/* bearer layer) is real, but loopback-binding only mitigates it — the WS path still isn't authenticated, which needs a server-side change.